Operators using AI hardest are pulling ahead. Data confirms.

The workers delegating the most work to AI aren't the ones worried about being replaced. They're the ones pulling ahead.

Anthropic published their June 2026 Economic Index last week. It's one of the cleaner primary-source data sets on how AI is actually changing work — real production conversation patterns, not sentiment surveys. The findings cut against most of the AI-displacement narrative circulating in SMB circles right now.

The counterintuitive part: the users who delegate the most tasks to AI are the most optimistic about their pay, job security, and career mobility. Not marginally more optimistic. Measurably. And the heaviest AI users by token volume aren't junior employees automating busy work — they're higher-wage roles. Marketing managers, operators, people with real production leverage.

If your ops team is using AI the same way they were eighteen months ago, that's not conservative. That's the risk position.

The data most ops teams haven't seen

The Economic Index analyzes actual Claude production conversations — token consumption across occupations, usage patterns, delegation levels. Not what people say they do with AI. What they actually do.

Here's what's already happening:

• 86% of heavy AI users report productivity speed gains. 82% report scope gains. 69% report quality gains.
• 57% say AI is making their skills more valuable, not obsolete.
• 68% say they're learning more because of AI use, not less.
• Higher-wage occupations generate roughly 2.5x more output tokens than lower-wage ones per session. Marketing managers at $80/hr generate roughly 2.5x the token volume of editors at $37/hr.

The pattern is consistent: the people using AI hardest are also the most optimistic about their professional future. That's not coincidence. That's compounding in real time.

Source: Anthropic — Economic Index June 2026 Report

One-third already expect AI to handle most of their tasks next year

Thirty-five percent of users already expect AI to handle "most or nearly all" of their work tasks within 12 months. Sixty percent expect materially more AI delegation than today.

These aren't people speculating about the future. These are users running AI at serious production volume. They know what it can take over because they've already handed it real work and watched what comes back.

The other two-thirds are describing AI the way people described email in 1995 — useful for some things, not central to how the work gets done. That framing worked for a while. Then it didn't.

The gap between "using AI sometimes" and "AI handles most of my production work" isn't a tool gap. It's a workflow design gap. And it widens every quarter you don't close it.

The workers using AI hardest aren't being replaced. They're setting the new baseline everyone else gets measured against.

Higher-wage roles aren't being displaced. They're doing more.

The standard AI fear is: the higher your market wage, the more exposed you are. The Economic Index data says the opposite.

Higher-wage occupations produce 1.34x more output per AI interaction than lower-wage ones. The $80/hour role isn't getting displaced by the model — it's using the model to generate 2.5x the production per session.

The model isn't replacing their judgment. It's multiplying how much they can apply per hour.

For SMB founders, the implication is specific: if your higher-value operators aren't actively compounding through AI, they're leaving leverage on the table — and they're falling behind peers at other companies running the same tools at 2.5x the intensity.

OpenAI's internal data makes the same case from a different angle: 80.6% of Codex requests represent tasks that would take a human more than 30 minutes. Their own workforce — the people who built the product — uses agents for work categories, not one-off asks. [1][2]

The compounding isn't theoretical. It's the operating model at the companies shipping these tools.

What serious AI use actually looks like

"We use AI" covers a lot of ground. Most of it is on the wrong end of the spectrum.

Casual AI use means: a ChatGPT tab open for writing help, occasional research queries, spot checks on drafts. Produces marginal productivity gains. Doesn't compound.

Serious AI use means: AI handles the first pass on entire categories of production work. Humans QA, direct, and handle the judgment calls. The operator's time concentrates on what only they can decide.

Here's the decision flow for where your team sits:

flowchart TD
  Start([Your ops team's AI posture]) --> Q1{Does AI handle full<br/>task categories before<br/>human review?}
  Q1 -->|Yes| TopThird[Top third:<br/>AI handles categories]
  Q1 -->|No| Q2{Does AI touch most<br/>production tasks in<br/>some form?}
  Q2 -->|Yes| MiddleHalf[Middle half:<br/>directionally right,<br/>not compounding yet]
  Q2 -->|No| Bottom[Bottom third:<br/>AI as convenience tool,<br/>not workflow layer]
  TopThird --> Outcome1["86% speed gains<br/>57% skill value up<br/>Quarterly compound"]
  MiddleHalf --> Outcome2["Marginal gains<br/>No compounding yet<br/>Gap closing vs. top"]
  Bottom --> Outcome3["Status quo<br/>Benchmark moves away<br/>Redesign needed"]

Serious AI use also requires someone owning the prompts, outputs, and iteration — not just submitting requests and accepting whatever comes back. That's the part most teams skip. They install a tool license, call it done, and wonder why the productivity gain is marginal.

The workflow has to be designed for AI participation, not retrofitted after the fact. That design work is where the compounding starts. It's also why the ops ceiling patterns I wrote about in June tend to show up before the productivity gap is visible in the numbers — the workflow debt accumulates quietly.

The test for your ops team

Here's the diagnostic: what percentage of your team's production hours does AI touch before a human reviews the output?

| Usage tier | What AI handles | Output multiple | Quarterly trajectory | |---|---|---|---| | Systematic (top third) | Full categories — drafts, research, reports, first-pass triage | ~2.5x vs. median | Compounding — gap widens in their favor | | Moderate (middle) | Some tasks, ad hoc — writing help, spot research, one-off queries | ~1.3x | Flat — marginal gains, no structural change | | Casual (bottom third) | Convenience use — grammar checks, quick summaries | ~1.0x | Falling behind — benchmark moves away quarterly |

If your answer to the diagnostic is "we use AI for writing help sometimes," you're in the bottom half.

If it's "AI handles first drafts of client briefs and reports, we QA and redirect," you're approaching the middle.

If it's "most of the routine production work runs through AI before a human sees it," you're in the cohort the Economic Index describes as the most productive, most optimistic, and most skill-amplified.

The test isn't philosophical. Count the task categories. Count which ones AI touches before a human does. If the number is small, the workflow wasn't designed for AI — and that's the redesign worth doing.

So what do you do with this

The Economic Index doesn't tell you what AI will do in five years. It tells you what the operators who got serious twelve months ago are experiencing right now. The gap is already there. It widens every quarter where the workflow design doesn't change.

If your ops team is somewhere in the bottom half of that table, the fix isn't a new tool license. It's workflow redesign — figuring out which task categories AI owns, how human review is structured, and who maintains the prompts. Specific, bounded work. Not a transformation project.

It's the kind of audit I run for fractional digital operator engagements — the same process behind the stack redesign work at The Hub. Not a pitch deck. A diagnostic, then a plan for the pieces worth executing.

Sources

[1] Anthropic — Economic Index June 2026 Report — https://www.anthropic.com/research/economic-index-june-2026-report [2] OpenAI — How Agents Are Transforming Work — https://openai.com/index/how-agents-are-transforming-work

The short version

• Heavy AI users report 86% speed gains, 82% scope gains, 69% quality gains — already happening, not projected
• 35% of users expect AI to handle most tasks within 12 months; 60% expect materially more delegation
• Higher-wage operators generate ~2.5x more output tokens than lower-wage peers — they're multiplying leverage, not being displaced
• The gap between casual AI use and systematic AI workflows compounds every quarter
• The diagnostic: what percentage of your team's production work does AI touch before a human? That number tells you which tier you're in